Proxy CommandCode

A transparent reverse proxy that translates OpenAI-compatible requests (/v1/chat/completions) into CommandCode's internal endpoint (/alpha/generate), then returns responses in OpenAI format your editor understands.

Use any CommandCode model (including the Go plan) with ZCode, 9router, Cursor, Continue, Aider, and any editor that supports custom OpenAI endpoints.

Fork changes (seankoji-com)

Forked from nasrulhadi/proxy-commandcode at 1a95fbb to run as a LiteLLM sidecar in a container. Differences:

Area	Change
Reasoning	Streamed and returned as reasoning_content, not mixed into content
Errors	An upstream error event before any output returns HTTP 502 (so a router can fail over); later events are ignored instead of writing to an ended response; network errors carry the code/addresses instead of an empty message
Usage	finish-step/finish usage mapped to OpenAI usage (sync body; extra stream chunk with empty choices)
Finish reason	length / tool_calls / content_filter mapped from the upstream reason
Request shape	Tool-call input sent as an object; consecutive tool results merged; developer role folded into system; image parts use {image, mediaType}; temperature and max_completion_tokens passed through; x-session-id and x-cli-environment headers sent
System prompt	A request with no (or a blank) system message gets a one-line default (PCMC_DEFAULT_SYSTEM, default You are a helpful assistant.); otherwise the gateway injects the ~7.3K-token cmd agent prompt on every call
CLI version	x-command-code-version tracks command-code@latest on npm (checked at start and every 6 h, default 1.66.0); set PCMC_VERSION to pin
Runtime	Logs to stdout only (no proxy.log, no ANSI without a TTY); Windows netstat/taskkill port handling removed; graceful SIGTERM; Dockerfile + npm test

Image: ghcr.io/seankoji-com/proxy-commandcode:sha-<commit>, built by .github/workflows/image.yml on every branch push. Deploy by digest.

Why

CommandCode has two API surfaces:

Endpoint	Plan required
/provider/v1/chat/completions	Provider plan
/provider/v1/messages	Provider plan
/alpha/generate	Any plan (Go included)

This proxy speaks the same envelope the official CLI uses, so it works on any plan.

How it works

Editor → POST /v1/chat/completions (OpenAI format)
  → Proxy translates to CommandCode format
    → POST api.commandcode.ai/alpha/generate
  ← Proxy translates NDJSON stream back to OpenAI SSE/JSON
← Editor receives standard OpenAI response

Quick start

git clone https://github.com/nasrulhadi/proxy-commandcode.git
cd proxy-commandcode
node server.js
# → listening on http://localhost:3456

Zero dependencies. Node.js 18+ only.

Get your token

1. Go to https://commandcode.ai/settings/billing
2. Copy your user_... token
3. Use it as the API key in your editor

Env vars

Variable	Default
PCMC_PORT	3456
PCMC_VERSION	unset — tracks npm command-code@latest (floor 1.66.0); set to pin
PCMC_DEBUG	off (set 1 to enable)

Enabling debug mode

PowerShell:

$env:PCMC_DEBUG="1"; node server.js

Command Prompt:

set PCMC_DEBUG=1 && node server.js

macOS / Linux / WSL:

PCMC_DEBUG=1 node server.js

When enabled, raw upstream responses are dumped to dump/ for troubleshooting.

Integration

9router

{
  "commandcode": {
    "base_url": "http://localhost:3456/v1",
    "api_key": "user_xxxxxxxxxx",
    "models": ["deepseek/deepseek-v4-pro", "moonshotai/Kimi-K2.5"]
  }
}

ZCode

Set custom OpenAI endpoint in settings:

• Base URL: http://localhost:3456/v1
• API Key: your user_... token
• Model: deepseek/deepseek-v4-pro

Cursor

Settings → OpenAI API Key → Override Base URL: http://localhost:3456/v1

Continue (VS Code)

{
  "models": [{
    "title": "DeepSeek V4 Pro",
    "provider": "openai",
    "apiBase": "http://localhost:3456/v1",
    "apiKey": "user_xxxxxxxxxx",
    "model": "deepseek/deepseek-v4-pro"
  }]
}

Aider

aider --openai-api-base http://localhost:3456/v1 \
      --openai-api-key user_xxxxxxxxxx \
      --model openai/deepseek/deepseek-v4-pro

Other editors

Any editor with custom OpenAI endpoint support works the same way: point base URL to http://localhost:3456/v1, use your token as API key.

Available models

Model availability depends on your CommandCode plan:

Plan	Price/mo	Access
Go	$1	Open-source + free + some premium (Luna, Grok 4.5, Muse Spark Contributor)
GOAT	$10	Go + per-model premium allowances (Sol, Grok 4.6, Muse Spark, Gemini 3.7/3.8)
Pro	$20	GOAT + full premium (GPT-6, Claude, Gemini)

Free models (cost no credits):

Model	Notes
poolside/laguna-s-2.1-free	Free while capacity lasts
meituan/LongCat-2.0:free	100 req/day · free while it lasts
inclusionai/ling-3.0-flash-sante:free	100 req/day · free while it lasts

Open-source — Go plan and above:

Model	Notes
deepseek/deepseek-v4-pro	Peak/off-peak pricing
deepseek/deepseek-v4-flash	Peak/off-peak pricing
deepseek/deepseek-v4-flash-fast
deepseek/deepseek-v4-flash-vision-exp
deepseek/deepseek-v4.1-flash	Boosted credits on GOAT/Pro until 2026-09-20
moonshotai/Kimi-K3
moonshotai/Kimi-K2.7-Code
moonshotai/Kimi-K2.7-Code-Highspeed
moonshotai/Kimi-K2.6
moonshotai/Kimi-K2.5
zai-org/GLM-5.3
z-ai/glm-5.3-flash
zai-org/GLM-5.2
zai-org/GLM-5.2-Fast
zai-org/GLM-5.1
zai-org/GLM-5
MiniMaxAI/MiniMax-M3	50% off
MiniMaxAI/MiniMax-M2.7
MiniMaxAI/MiniMax-M2.5
xiaomi/mimo-v2.5-pro	Up to 99% off
xiaomi/mimo-v2.5	Up to 98% off
Qwen/Qwen3.8-Max-0902
Qwen/Qwen3.8-Max
Qwen/Qwen3.8-Flash
Qwen/Qwen3.8-27B
Qwen/Qwen3.7-Max
Qwen/Qwen3.7-Plus
Qwen/Qwen3.7-Flash
Qwen/Qwen3.6-Max-Preview
Qwen/Qwen3.6-Plus
stepfun/Step-3.7-Flash
stepfun/Step-3.5-Flash
tencent/hy4-preview
tencent/hy3-paid
nvidia/nemotron-3-ultra-550b-a55b
thinkingmachines/inkling
thinkingmachines/inkling-small

Premium — Go plan and above:

Model	Notes
gpt-5.6-luna	Not available in China
xai/grok-4.5
meta/muse-spark-1.2-contributor	~95% off
meta/muse-spark-1.3-contributor	Up to 95% off

GOAT plan and above:

Model	Notes
gpt-5.6-sol
xai/grok-4.6
meta/muse-spark-1.2
meta/muse-spark-1.3
google/gemini-3.7-flash	Not available in China
google/gemini-3.8-flash

Premium — Pro plan and above:

Model	Notes
gpt-6-astra
gpt-5.6-terra
gpt-5.5
gpt-5.4
gpt-5.4-mini
gpt-5.3-codex
claude-fable-5-1
claude-fable-5
claude-opus-5
claude-opus-4-8
claude-opus-4-7
claude-sonnet-5
claude-sonnet-4-6
claude-haiku-4-5
google/gemini-3.6-flash
google/gemini-3.5-flash
google/gemini-3.5-flash-lite
google/gemini-3.1-flash-lite
meta/muse-spark-1.1
sakana/fugu-ultra

DeepSeek V4 models: peak rates apply 01:00–04:00 and 06:00–10:00 UTC Mon–Fri; everything else, plus full weekends, is half price.

See all models → cmd --list-models for the current roster.

Logs

All requests are logged to both terminal (colorized) and proxy.log (plain text):

[2026-07-09T09:10:06.092Z] [req] tencent/Hy3 | ::1 | sync | 232 bytes
[2026-07-09T09:10:07.137Z] [upstream] 200 OK | tencent/Hy3 | 1045ms
[2026-07-09T09:10:09.303Z] [done] tencent/Hy3 | 92 text / 0 tools | stop | 3211ms

Color legend (terminal only):

• [req] — cyan — incoming request with model, client IP, stream mode, byte size
• [upstream] — green — HTTP status, model, latency to CommandCode
• [done] — bold — text / reasoning / tool counts, finish reason, total duration
• [error] — yellow — upstream or internal errors

Log fields:

• text / reasoning — total characters of each type in the response
• tools — number of tool calls the model made
• reason — stop (finished) or tool_calls (waiting for tool results)

Common errors

Log	Fix
[upstream] 401	Token invalid. Get a fresh one from billing.
[upstream] 400	Request format mismatch. Check proxy version.
[upstream] timeout	Upstream took >5 min. Retry.
[error] context length	Prompt too large for model (e.g. ~520K tokens vs 256K max). Reduce messages or switch to a higher-context model.
EADDRINUSE	Proxy auto-kills the old process on start.

Disclaimer

This project is for educational and research purposes only. It uses CommandCode's internal /alpha/generate endpoint, which is undocumented and may change or be restricted at any time. Use at your own risk and in accordance with CommandCode's Terms of Service.

License

MIT